Repotrace helps AI agents track software needs through many steps
RepoTrace: Improving Requirement Tracing in Multi-Round Agent Interactions
Software EngineeringArtificial Intelligence
Summary
When AI programs modify software step by step, they often lose track of what parts to change. The authors found that AI agents struggle to focus on the main software requirements during long tasks. They created Repotrace, a tool that helps agents understand how software files relate to specific needs. This helps the AI keep better track across many steps and reduces errors during software updates.
What this means in practice
- •For software development teams: Improve AI-driven code modifications by maintaining clear links between requirements and code across iterative changes.
- •For software quality assurance teams: Use enhanced requirement tracing to reduce regression errors and increase reliability when AI agents handle software updates.
Authors
Huanyu Liu, Yezheng Cheng, Jinliang You, Tianxiao Huang, Linyu Li, Zheng Fang, Peng Cao, Ge Li
Abstract
Code agents have become increasingly capable of performing software engineering tasks through interaction with software environments. However, agents tend to degrade as the number of interaction rounds increases, making it difficult to maintain existing software during iterative modification. We analyze 2,656 agent trajectories in EvoCode-Bench and identify requirement focus misalignment as a recurring failure pattern, where agents struggle to distinguish the primary modification targets of a requirement from unrelated code. We propose Repotrace a requirement-driven repository understanding harness that combines repository structure with requirement semantics to construct a requirement-code correspondence and injects it into the agent context. This correspondence enables code agents to follow the repository architecture, trace requirements across multi-round interactions, and identify the core code elements corresponding to the current requirement. Experiments show that Repotrace improves the mean per-round reward (MPR) across different models (e.g., +19.99 on GLM-5.2) and agent harnesses (e.g., +10.22 on Claude Code). Across rounds, Repotrace increases the average per-round reward from 0.163 to 0.371 on Terminus-2 with DeepSeek-V4-Pro, while reducing average regression errors by 56.78%. We release the source code and agents' trajectories to support future research.